| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
With respect for the opinions of those who intensely disagree, I vote no:
1. 1788 should only specify what is to be accomplished not how things are done, so should not require Complete Arithmetic.
2. CA might not be appropriate in all cases, including for IEEE binary128 quad precision and extended precision formats.
It may not even be best for single and double precision performance.
3. This amendment reverses the intent of the original motion. Amendments should not do that.
http://www.dummies.com/how-to/content/roberts-rules-for-amending-a-motion.html says
"Another important rule to remember is that an amendment that does nothing but make the motion a rejection of the original motion is not proper and not in order. "
It should be a separate motion, not an amendment.
I would vote for the amendment if it did not require complete arithmetic.
It doesn't affect my vote, but FWIW:
1. CA should have its own standard, not be part of an Interval Arithmetic standard. (I have already volunteered to help with that.)
2. My understanding of 1788's mandate is to provide programmers and users of other tools with a way to know the error bounds of their calculations.
EDP and better math functions are worthy objectives, but aren't part of what I think our mandate is.
They should be part of the next 754 standard or have their own standards.
3. Trying to standardize too many things jeopardizes 1788's schedule.
- Ian McIntosh IBM Canada Lab Compiler Back End Support and Development